loading...
Encore AI mines what top employees do on customer calls – then trains AI agents to replicate that behavior at scale.
ENTRY ANGLES
Vertical-specific interaction mining for insurance claims adjusting · AI co-pilot for high-stakes healthcare call coordination · Interaction mining for legal intake qualification
VERTICALS
CAPABILITIES
Speech-to-text processing, CRM integration, ML training pipelines, Compliance expertise, Enterprise sales
Most enterprise voice AI is built to deflect. Encore AI is built to sell.
The distinction sounds tactical. It isn't. Deflection-optimized agents are designed around a single metric – cost per interaction – and they execute against it reliably. They route calls, field FAQs, and close out customer inquiries before a human ever gets involved. But every deflected interaction is a negotiation that didn't happen, a cross-sell that was never attempted, a retention conversation that defaulted to cancellation.
Encore AI – founded in 2022 as Insait IO and rebranded after its core thesis sharpened – inverts the standard workflow. Rather than scripting agent behavior from scratch, the company mines its clients' existing interaction archives: call recordings, emails, texts, and CRM notes. Its models identify which conversation patterns, response sequences, and handling approaches correlate with successful outcomes – the calls where customers upgraded, stayed, or deepened their investment. Those patterns become the training signal.
The process Encore calls "interaction mining" produces agents calibrated to what demonstrably worked, not what a product manager speculated would work. In deployment, the agents run in two modes: fully autonomous, handling interactions end-to-end, or as a real-time co-pilot riding alongside human reps – suggesting responses mid-call, flagging risk signals, recommending approaches drawn from the training set.
Financial services is the current market. Banks, insurers, and wealth management firms are a logical wedge: regulatory requirements mean calls are already recorded, interaction volumes run into millions annually, and the cost of a mishandled retention conversation is directly measurable in AUM. Encore has 40+ enterprise customers in this space; annualized revenue is up 5x since the seed round 18 months ago. The $30M Series A, led by Team8, included commercial banks and insurers that crossed from customers to investors.
Gong reached a $7.2 billion valuation by extracting insights from sales call recordings and routing them to coaching dashboards. Encore is making the adjacent but structurally different bet: that the same data can train the agent, not just advise the human.
The difference compounds. Coaching insights are static – the pattern observed in January is the recommended practice in December. An agent trained on interaction data improves as new interactions are ingested, and the improvement is specific to each client's conversation history, which competitors cannot replicate without that same data.
This creates a switching-cost dynamic worth understanding. The first enterprise that gives Encore access to its full interaction archive and allows the models to train on it is building something that accumulates. Replacing that deployment 18 months later requires finding a vendor with models that are better and who can reproduce the behavioral fine-tuning on equivalent historical data. The switching cost isn't the subscription; it's the accumulated training.
The broader industry bet: deflection metrics are being scrutinized. Several large financial services firms have disclosed that contact-center deflection improved their cost-per-call figures while simultaneously worsening their NPS scores. The correlation between deflection and customer lifetime value is negative in most high-consideration financial products. Encore's pitch is that someone was always going to offer an AI product optimized for the opposite direction – and that the data moat makes that product more defensible the earlier it's deployed.
Interaction mining applies wherever expertise is locked in conversation history and outcomes are quantifiably attached to those conversations.
Financial services was the right first market: compliance recording is already mandatory, volumes are enormous, and a mishandled call is quantifiable in AUM or premium terms. The same structure holds for insurance claims adjusting – calls between adjusters and policyholders, with outcomes measured as settlement amounts and retention rates. For legal intake – conversations between paralegals and potential clients, with outcomes as case acceptance rates. For medical appointment coordination – calls between care coordinators and patients, with outcomes as show rates and care-plan compliance.
What each of these has that makes Encore's model applicable: an existing recorded archive, measurable outcomes tied to interaction quality, and no existing vendor approaching the problem from "train on what worked" rather than scripting from scratch.
The constraint that defines the entry point is worth naming precisely: the archive has to exist before mining can begin. A company that recorded nothing has nothing to train on. The natural early-customer profile is large, regulated, interaction-heavy, and already recording. The natural laggard is younger, with sparse history. That asymmetry is the moat – and the first enterprise customer in any given vertical who allows model training on their full archive has a compounding head start that latecomers cannot easily close.